OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials.

Machine learning plays an important and growing role in molecular simulation. The newest version of the OpenMM molecular dynamics toolkit introduces new features to support the use of machine learning potentials. Arbitrary PyTorch models can be added to a simulation and used to compute forces and energy. A higher-level interface allows users to easily model their molecules of interest with general purpose, pretrained potential functions. A collection of optimized CUDA kernels and custom PyTorch operations greatly improves the speed of simulations. We demonstrate these features in simulations of cyclin-dependent kinase 8 (CDK8) and the green fluorescent protein chromophore in water. Taken together, these features make it practical to use machine learning to improve the accuracy of simulations with only a modest increase in cost.

OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials.

Machine learning plays an important and growing role in molecular simulation. The newest version of the OpenMM molecular dynamics toolkit introduces new features to support the use of machine learning potentials. Arbitrary PyTorch models can be added to a simulation and used to compute forces and energy. A higher-level interface allows users to easily model their molecules of interest with general purpose, pretrained potential functions. A collection of optimized CUDA kernels and custom PyTorch operations greatly improves the speed of simulations. We demonstrate these features on simulations of cyclin-dependent kinase 8 (CDK8) and the green fluorescent protein (GFP) chromophore in water. Taken together, these features make it practical to use machine learning to improve the accuracy of simulations at only a modest increase in cost.

Folding@home: Achievements from over 20 years of citizen science herald the exascale era.

Simulations of biomolecules have enormous potential to inform our understanding of biology but require extremely demanding calculations. For over 20 years, the Folding@home distributed computing project has pioneered a massively parallel approach to biomolecular simulation, harnessing the resources of citizen scientists across the globe. Here, we summarize the scientific and technical advances this perspective has enabled. As the project's name implies, the early years of Folding@home focused on driving advances in our understanding of protein folding by developing statistical methods for capturing long-timescale processes and facilitating insight into complex dynamical processes. Success laid a foundation for broadening the scope of Folding@home to address other functionally relevant conformational changes, such as receptor signaling, enzyme dynamics, and ligand binding. Continued algorithmic advances, hardware developments such as graphics processing unit (GPU)-based computing, and the growing scale of Folding@home have enabled the project to focus on new areas where massively parallel sampling can be impactful. While previous work sought to expand toward larger proteins with slower conformational changes, new work focuses on large-scale comparative studies of different protein sequences and chemical compounds to better understand biology and inform the development of small-molecule drugs. Progress on these fronts enabled the community to pivot quickly in response to the COVID-19 pandemic, expanding to become the world's first exascale computer and deploying this massive resource to provide insight into the inner workings of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus and aid the development of new antivirals. This success provides a glimpse of what is to come as exascale supercomputers come online and as Folding@home continues its work.

Folding@home: achievements from over twenty years of citizen science herald the exascale era.

Simulations of biomolecules have enormous potential to inform our understanding of biology but require extremely demanding calculations. For over twenty years, the Folding@home distributed computing project has pioneered a massively parallel approach to biomolecular simulation, harnessing the resources of citizen scientists across the globe. Here, we summarize the scientific and technical advances this perspective has enabled. As the project's name implies, the early years of Folding@home focused on driving advances in our understanding of protein folding by developing statistical methods for capturing long-timescale processes and facilitating insight into complex dynamical processes. Success laid a foundation for broadening the scope of Folding@home to address other functionally relevant conformational changes, such as receptor signaling, enzyme dynamics, and ligand binding. Continued algorithmic advances, hardware developments such as GPU-based computing, and the growing scale of Folding@home have enabled the project to focus on new areas where massively parallel sampling can be impactful. While previous work sought to expand toward larger proteins with slower conformational changes, new work focuses on large-scale comparative studies of different protein sequences and chemical compounds to better understand biology and inform the development of small molecule drugs. Progress on these fronts enabled the community to pivot quickly in response to the COVID-19 pandemic, expanding to become the world's first exascale computer and deploying this massive resource to provide insight into the inner workings of the SARS-CoV-2 virus and aid the development of new antivirals. This success provides a glimpse of what's to come as exascale supercomputers come online, and Folding@home continues its work.

Improvement in ADMET Prediction with Multitask Deep Featurization.

The absorption, distribution, metabolism, elimination, and toxicity (ADMET) properties of drug candidates are important for their efficacy and safety as therapeutics. Predicting ADMET properties has therefore been of great interest to the computational chemistry and medicinal chemistry communities in recent decades. Traditional cheminformatics approaches, using learners such as random forests and deep neural networks, leverage fingerprint feature representations of molecules. Here, we learn the features most relevant to each chemical task at hand by representing each molecule explicitly as a graph. By applying graph convolutions to this explicit molecular representation, we achieve, to our knowledge, unprecedented accuracy in prediction of ADMET properties. By challenging our methodology with rigorous cross-validation procedures and prognostic analyses, we show that deep featurization better enables molecular predictors to not only interpolate but also extrapolate to new regions of chemical space.

Dynamical model of the CLC-2 ion channel reveals conformational changes associated with selectivity-filter gating.

This work reports a dynamical Markov state model of CLC-2 "fast" (pore) gating, based on 600 microseconds of molecular dynamics (MD) simulation. In the starting conformation of our CLC-2 model, both outer and inner channel gates are closed. The first conformational change in our dataset involves rotation of the inner-gate backbone along residues S168-G169-I170. This change is strikingly similar to that observed in the cryo-EM structure of the bovine CLC-K channel, though the volume of the intracellular (inner) region of the ion conduction pathway is further expanded in our model. From this state (inner gate open and outer gate closed), two additional states are observed, each involving a unique rotameric flip of the outer-gate residue GLUex. Both additional states involve conformational changes that orient GLUex away from the extracellular (outer) region of the ion conduction pathway. In the first additional state, the rotameric flip of GLUex results in an open, or near-open, channel pore. The equilibrium population of this state is low (∼1%), consistent with the low open probability of CLC-2 observed experimentally in the absence of a membrane potential stimulus (0 mV). In the second additional state, GLUex rotates to occlude the channel pore. This state, which has a low equilibrium population (∼1%), is only accessible when GLUex is protonated. Together, these pathways model the opening of both an inner and outer gate within the CLC-2 selectivity filter, as a function of GLUex protonation. Collectively, our findings are consistent with published experimental analyses of CLC-2 gating and provide a high-resolution structural model to guide future investigations.

A modular DNA scaffold to study protein-protein interactions at single-molecule resolution.

The residence time of a drug on its target has been suggested as a more pertinent metric of therapeutic efficacy than the traditionally used affinity constant. Here, we introduce junctured-DNA tweezers as a generic platform that enables real-time observation, at the single-molecule level, of biomolecular interactions. This tool corresponds to a double-strand DNA scaffold that can be nanomanipulated and on which proteins of interest can be engrafted thanks to widely used genetic tagging strategies. Thus, junctured-DNA tweezers allow a straightforward and robust access to single-molecule force spectroscopy in drug discovery, and more generally in biophysics. Proof-of-principle experiments are provided for the rapamycin-mediated association between FKBP12 and FRB, a system relevant in both medicine and chemical biology. Individual interactions were monitored under a range of applied forces and temperatures, yielding after analysis the characteristic features of the energy profile along the dissociation landscape.

Synthesis of spiro-2,6-dioxopiperazine and spiro-2,6-dioxopyrazine scaffolds using amino acids in a three-component reaction to generate potential Sigma-1 (σ1) receptor selective ligands.

A library-friendly approach to generate new scaffolds is decisive for the development of molecular probes, drug like molecules and preclinical entities. Here, we present the design and synthesis of novel heterocycles with spiro-2,6-dioxopiperazine and spiro-2,6-pyrazine scaffolds through a three-component reaction using various amino acids, ketones, and isocyanides. Screening of select compounds over fifty CNS receptors including G-protein coupled receptors (GPCRs), ion channels, transporters, and enzymes through the NIMH psychoactive drug screening program indicated that a novel spiro-2,6-dioxopyrazine scaffold, UVM147, displays high binding affinity at sigma-1 (σ1) receptor in the nanomolar range. In addition, molecular docking of UVM147 at the human σ1 receptor have shown that it resides in the same binding site that was occupied by the ligand 4-IBP used to obtain a crystal structure of the human sigma-1 (σ1) receptor.

Intradomain Interactions in an NMDA Receptor Fragment Mediate N-Glycan Processing and Conformational Sampling.

The structural and functional roles of highly conserved asparagine-linked (N)-glycans on the extracellular ligand-binding domain (LBD) of the N-methyl-D-aspartate receptors are poorly understood. We applied solution- and computation-based methods that identified N-glycan-mediated intradomain and interglycan interactions. Nuclear magnetic resonance (NMR) spectra of the GluN1 LBD showed clear signals corresponding to each of the three N-glycans and indicated the reducing end of glycans at N440 and N771 potentially contacted nearby amino acids. Molecular dynamics simulations identified contacts between nearby amino acids and the N440- and N771-glycans that were consistent with the NMR spectra. The distal portions of the N771-glycan also contacted the core residues of the nearby N471-glycan. This result was consistent with mass spectrometry data indicating the limited N471-glycan core fucosylation and reduced branch processing of the N771-glycan could be explained by interglycan contacts. We discuss a potential role for the GluN1 LBD N-glycans in interdomain contacts formed in NMDA receptors.

Note: Variational encoding of protein dynamics benefits from maximizing latent autocorrelation.

As deep Variational Auto-Encoder (VAE) frameworks become more widely used for modeling biomolecular simulation data, we emphasize the capability of the VAE architecture to concurrently maximize the time scale of the latent space while inferring a reduced coordinate, which assists in finding slow processes as according to the variational approach to conformational dynamics. We provide evidence that the VDE framework [Hernández et al., Phys. Rev. E 97, 062412 (2018)], which uses this autocorrelation loss along with a time-lagged reconstruction loss, obtains a variationally optimized latent coordinate in comparison with related loss functions. We thus recommend leveraging the autocorrelation of the latent space while training neural network models of biomolecular simulation data to better represent slow processes.

PotentialNet for Molecular Property Prediction.

The arc of drug discovery entails a multiparameter optimization problem spanning vast length scales. The key parameters range from solubility (angstroms) to protein-ligand binding (nanometers) to in vivo toxicity (meters). Through feature learning-instead of feature engineering-deep neural networks promise to outperform both traditional physics-based and knowledge-based machine learning models for predicting molecular properties pertinent to drug discovery. To this end, we present the PotentialNet family of graph convolutions. These models are specifically designed for and achieve state-of-the-art performance for protein-ligand binding affinity. We further validate these deep neural networks by setting new standards of performance in several ligand-based tasks. In parallel, we introduce a new metric, the Regression Enrichment Factor EFχ (R), to measure the early enrichment of computational models for chemical data. Finally, we introduce a cross-validation strategy based on structural homology clustering that can more accurately measure model generalizability, which crucially distinguishes the aims of machine learning for drug discovery from standard machine learning tasks.

Perspective: Computational chemistry software and its advancement as illustrated through three grand challenge cases for molecular science.

The field of computational molecular sciences (CMSs) has made innumerable contributions to the understanding of the molecular phenomena that underlie and control chemical processes, which is manifested in a large number of community software projects and codes. The CMS community is now poised to take the next transformative steps of better training in modern software design and engineering methods and tools, increasing interoperability through more systematic adoption of agreed upon standards and accepted best-practices, overcoming unnecessary redundancy in software effort along with greater reproducibility, and increasing the deployment of new software onto hardware platforms from in-house clusters to mid-range computing systems through to modern supercomputers. This in turn will have future impact on the software that will be created to address grand challenge science that we illustrate here: the formulation of diverse catalysts, descriptions of long-range charge and excitation transfer, and development of structural ensembles for intrinsically disordered proteins.

Bayesian analysis of isothermal titration calorimetry for binding thermodynamics.

Isothermal titration calorimetry (ITC) is the only technique able to determine both the enthalpy and entropy of noncovalent association in a single experiment. The standard data analysis method based on nonlinear regression, however, provides unrealistically small uncertainty estimates due to its neglect of dominant sources of error. Here, we present a Bayesian framework for sampling from the posterior distribution of all thermodynamic parameters and other quantities of interest from one or more ITC experiments, allowing uncertainties and correlations to be quantitatively assessed. For a series of ITC measurements on metal:chelator and protein:ligand systems, the Bayesian approach yields uncertainties which represent the variability from experiment to experiment more accurately than the standard data analysis. In some datasets, the median enthalpy of binding is shifted by as much as 1.5 kcal/mol. A Python implementation suitable for analysis of data generated by MicroCal instruments (and adaptable to other calorimeters) is freely available online.

Automated design of collective variables using supervised machine learning.

Selection of appropriate collective variables (CVs) for enhancing sampling of molecular simulations remains an unsolved problem in computational modeling. In particular, picking initial CVs is particularly challenging in higher dimensions. Which atomic coordinates or transforms there of from a list of thousands should one pick for enhanced sampling runs? How does a modeler even begin to pick starting coordinates for investigation? This remains true even in the case of simple two state systems and only increases in difficulty for multi-state systems. In this work, we solve the "initial" CV problem using a data-driven approach inspired by the field of supervised machine learning (SML). In particular, we show how the decision functions in SML algorithms can be used as initial CVs (SMLcv ) for accelerated sampling. Using solvated alanine dipeptide and Chignolin mini-protein as our test cases, we illustrate how the distance to the support vector machines' decision hyperplane, the output probability estimates from logistic regression, the outputs from shallow or deep neural network classifiers, and other classifiers may be used to reversibly sample slow structural transitions. We discuss the utility of other SML algorithms that might be useful for identifying CVs for accelerating molecular simulations.

Engineering a Single-Agent Cytokine/Antibody Fusion That Selectively Expands Regulatory T Cells for Autoimmune Disease Therapy.

IL-2 has been used to treat diseases ranging from cancer to autoimmune disorders, but its concurrent immunostimulatory and immunosuppressive effects hinder efficacy. IL-2 orchestrates immune cell function through activation of a high-affinity heterotrimeric receptor (composed of IL-2Rα, IL-2Rß, and common γ [γc]). IL-2Rα, which is highly expressed on regulatory T (TReg) cells, regulates IL-2 sensitivity. Previous studies have shown that complexation of IL-2 with the JES6-1 Ab preferentially biases cytokine activity toward TReg cells through a unique mechanism whereby IL-2 is exchanged from the Ab to IL-2Rα. However, clinical adoption of a mixed Ab/cytokine complex regimen is limited by stoichiometry and stability concerns. In this study, through structure-guided design, we engineered a single agent fusion of the IL-2 cytokine and JES6-1 Ab that, despite being covalently linked, preserves IL-2 exchange, selectively stimulating TReg expansion and exhibiting superior disease control to the mixed IL-2/JES6-1 complex in a mouse colitis model. These studies provide an engineering blueprint for resolving a major barrier to the implementation of functionally similar IL-2/Ab complexes for treatment of human disease.

Computer Simulations Predict High Structural Heterogeneity of Functional State of NMDA Receptors.

N-methyl-D-aspartate receptors (NMDARs)-i.e., transmembrane proteins expressed in neurons-play a central role in the molecular mechanisms of learning and memory formation. It is unclear how the known atomic structures of NMDARs determined by x-ray crystallography and electron cryomicroscopy (18 published Protein Data Bank entries) relate to the functional states of NMDARs inferred from electrophysiological recordings (multiple closed, open, preopen, etc. states). We address this problem by using molecular dynamics simulations at atomic resolution, a method successfully applied in the past to much smaller biomolecules. Our simulations predict that several conformations of NMDARs with experimentally determined geometries, including four "nonactive" electron cryomicroscopy structures, rapidly interconvert on submicrosecond timescales and therefore may correspond to the same functional state of the receptor (specifically, one of the closed states). This conclusion is not trivial because these conformational transitions involve changes in certain interatomic distances as large as tens of Å. The simulations also predict differences in the conformational dynamics of the apo and holo (i.e., agonist and coagonist bound) forms of the receptor on the microsecond timescale. To our knowledge, five new conformations of NMDARs, with geometries joining various features from different known experimental structures, are also predicted by the model. The main limitation of this work stems from its limited sampling (30 µs of aggregate length of molecular dynamics trajectories). Though this level significantly exceeds the sampling in previous simulations of parts of NMDARs, it is still much lower than the sampling recently achieved for smaller biomolecules (up to a few milliseconds), thus precluding, in particular, the observation of transitions between different functional states of NMDARs. Despite this limitation, such computational predictions may guide further experimental studies on the structure, dynamics, and function of NMDARs, for example by suggesting optimal locations of spectroscopic probes. Overall, atomic resolution simulations provide, to our knowledge, a novel perspective on the structure and dynamics of NMDARs, complementing information obtained by experimental methods.

Towards simple kinetic models of functional dynamics for a kinase subfamily.

Kinases are ubiquitous enzymes involved in the regulation of critical cellular pathways. However, in silico modelling of the conformational ensembles of these enzymes is difficult due to inherent limitations and the cost of computational approaches. Recent algorithmic advances combined with homology modelling and parallel simulations have enabled researchers to address this computational sampling bottleneck. Here, we present the results of molecular dynamics studies for seven Src family kinase (SFK) members: Fyn, Lyn, Lck, Hck, Fgr, Yes and Blk. We present a sequence invariant extension to Markov state models, which allows us to quantitatively compare the structural ensembles of the seven kinases. Our findings indicate that in the absence of their regulatory partners, SFK members have similar in silico dynamics with active state populations ranging from 4 to 40% and activation timescales in the hundreds of microseconds. Furthermore, we observe several potentially druggable intermediate states, including a pocket next to the adenosine triphosphate binding site that could potentially be targeted via a small-molecule inhibitor.

Variational encoding of complex dynamics.

Often the analysis of time-dependent chemical and biophysical systems produces high-dimensional time-series data for which it can be difficult to interpret which individual features are most salient. While recent work from our group and others has demonstrated the utility of time-lagged covariate models to study such systems, linearity assumptions can limit the compression of inherently nonlinear dynamics into just a few characteristic components. Recent work in the field of deep learning has led to the development of the variational autoencoder (VAE), which is able to compress complex datasets into simpler manifolds. We present the use of a time-lagged VAE, or variational dynamics encoder (VDE), to reduce complex, nonlinear processes to a single embedding with high fidelity to the underlying dynamics. We demonstrate how the VDE is able to capture nontrivial dynamics in a variety of examples, including Brownian dynamics and atomistic protein folding. Additionally, we demonstrate a method for analyzing the VDE model, inspired by saliency mapping, to determine what features are selected by the VDE model to describe dynamics. The VDE presents an important step in applying techniques from deep learning to more accurately model and interpret complex biophysics.

Solving the RNA design problem with reinforcement learning.

We use reinforcement learning to train an agent for computational RNA design: given a target secondary structure, design a sequence that folds to that structure in silico. Our agent uses a novel graph convolutional architecture allowing a single model to be applied to arbitrary target structures of any length. After training it on randomly generated targets, we test it on the Eterna100 benchmark and find it outperforms all previous algorithms. Analysis of its solutions shows it has successfully learned some advanced strategies identified by players of the game Eterna, allowing it to solve some very difficult structures. On the other hand, it has failed to learn other strategies, possibly because they were not required for the targets in the training set. This suggests the possibility that future improvements to the training protocol may yield further gains in performance.

Communication: Adaptive boundaries in multiscale simulations.

Combined-resolution simulations are an effective way to study molecular properties across a range of length and time scales. These simulations can benefit from adaptive boundaries that allow the high-resolution region to adapt (change size and/or shape) as the simulation progresses. The number of degrees of freedom required to accurately represent even a simple molecular process can vary by several orders of magnitude throughout the course of a simulation, and adaptive boundaries react to these changes to include an appropriate but not excessive amount of detail. Here, we derive the Hamiltonian and distribution function for such a molecular simulation. We also design an algorithm that can efficiently sample the boundary as a new coordinate of the system. We apply this framework to a mixed explicit/continuum simulation of a peptide in solvent. We use this example to discuss the conditions necessary for a successful implementation of adaptive boundaries that is both efficient and accurate in reproducing molecular properties.